Developing Autopilot Agent Transparency for Collaborative Driving

Collaborative driving is considered as a form of human-autonomy teaming (HAT) in which the advanced driving assistance system (ADAS) with an autopilot feature plays a role as the human driver counterpart, not merely as an automation tool. However, such a collaborative driving raises a problem for the human driver's situational awareness development, particularly because of the lack of mechanisms to comprehend the autopilot agent's behaviours. The human driver becomes overly trust to the agent and is vulnerable to distractions. As a result, many road incidents occur because of such mental model. It is believed that the transparency of the autopilot agent can help its human counterpart to calibrate their trust in this agent. However, a lack of studies investigating how such transparency is delivered to the human driver. Hence, this study aims to develop autopilot agent transparency for collaborative driving. The developed transparency is implemented and simulated using open-source software for autonomous driving called Carla simulator. The findings show that the transparency can help the human driver to understand and predict the autopilot agent's behaviours better. Such transparency is critical to enhance human-machine interaction, particularly in a collaborative driving context.